policyengine/microcosm-evaluation
Microcosm cross-dataset evaluation bundles Immutable frontend bundles (cross_dataset.frontend_bundle.v1) consumed by the calibration diagnostics dashboard's Cross-dataset page: every source (a model × dataset pair, or a standalone dataset) scored against the same Chronicle facts for a jurisdiction. Produced by PolicyEngine/calibration-diagnostics scripts/run_full_chronicle_evaluation.py + publish_cross_dataset_frontend_bundle.py; every partition carries the run and snapshot ids… See the full description on the dataset page: https://huggingface.co/datasets/policyengine/microcosm-evaluation.
Microcosm cross-dataset evaluation bundles
Immutable frontend bundles (cross_dataset.frontend_bundle.v1) consumed by the calibration diagnostics dashboard's Cross-dataset page: every source (a model × dataset pair, or a standalone dataset) scored against the same Chronicle facts for a jurisdiction. Produced by PolicyEngine/calibration-diagnostics scripts/run_full_chronicle_evaluation.py + publish_cross_dataset_frontend_bundle.py; every partition carries the run and snapshot ids and a SHA-256 recorded in manifest.json.
Layout: <jurisdiction>/<run_id>/frontend/ and <jurisdiction>/latest.json → run id.
Belgium (evaluation-f28ca06a0b0d2baf13c87f2f): sources Microcosm-BE v0.4 × Axiom rules engine, Microcosm-BE v0.4 × EUROMOD BE2025, EUROMOD BE2025 on EU-SILC (JRC country report). Microcosm-BE provenance: synthetic Belgian population calibrated to Belgian administrative and national-accounts targets (sums of Chronicle facts; surveys validation-only); support records are a US survey donor pool, reweighted; a Belgian donor pool is the planned upgrade.